Diagnosis of plant diseases with artificial neural networks

As with all diseases, early detection is important in crops. Progressive disease or damage results in the loss of the plant, the spread of the disease to neighboring fields or the use of more pesticides to eradicate the disease. This leads to economic losses in the agricultural sector, reduced yields and negative impacts on the health of end-users due to over-spraying. Plant diseases and pests cause localized visual changes in the leaves or fruits of plants. Nowadays, machine learning models utilize these visual changes to detect and diagnose diseases or pests. In this study, Artificial Neural Network models and methods are reviewed and experimental studies are conducted to improve the performance of the models trained with a small number of diseased plant image datasets. A dataset was created by collecting images showing the damages caused by leafminer fly on lettuce plants selected as a sample plant. With the lettuce plant training set, Convolutional Neural Network models created by both scratch learning and transfer learning were trained. The trained base models were tested with lettuce test dataset and tomato, potato and zucchini test datasets and their performances were compared. Ensemble learning methods such as boosting, bagging, stacking and voting were used to improve the performance of the base models. As a result of the experiments, it is shown that the combination of Convolutional Neural Networks and ensemble learning models gives better results than the baseline models.